Developing dynamic structure-based pharmacophore and QSAR models for the discovery of novel resistance-free RET tyrosine kinase inhibitors through extensive molecular dynamics trajectories for accelerated and successful virtual screening of ultra large ligand libraries

Activation of RET tyrosine kinase plays a critical role in the pathogenesis of various cancers, including non-small cell lung cancer, papillary thyroid cancers, multiple endocrine neoplasia type 2A and 2B (MEN2A, MEN2B), and familial medullary thyroid cancer. Gene fusions and point mutations in the RET proto-oncogene result in constitutive activation of RET signaling pathways. Consequently, developing effective inhibitors to target RET is of utmost importance. Small molecules have shown promise as inhibitors by binding to the kinase domain of RET and blocking its enzymatic activity. However, the emergence of resistance due to single amino acid changes poses a significant challenge. In this study, trajectories recorded during the molecular dynamics (MD) simulations are used to develop dynamic structure-based pharmacophore models. The developed pharmacophore hypotheses are then used in deep learning approaches and evaluated with neural relational inference (NRI) and density-based spatial clustering of applications with noise-(DBSCAN) analyses. The NRI and DBSCAN methods are used to analyze relationships between pharmacophore features based on MD simulation trajectory data. ML-trained QSAR models were also developed to predict pIC50 values of compounds. For this aim, extensive small molecule libraries were screened using developed hybrid-based models and potent compounds which are capable of inhibiting RET activation were proposed.